AMD Announcement: ARMv8 Opterons In 2014

[ViRGE]

Well, I keep seeing a lot of really bad assumptions that arm has *any* server presence at all. I guess it being my backyard, I am more up on it, but it boggles my mind just how widespread a few buzzword happy articles on the web can mislead people.

You're probably right. I'm not saying that this is going to work or that ARM makes sense in servers. I'm only saying that ARM in general right now is a growing market, and servers will be their next push. AMD feels that they need to do something new, and based on ARM growth they think this will be the emerging market that pays off for them.

 

[Ferzerp]

You're probably right. I'm not saying that this is going to work or that ARM makes sense in servers. I'm only saying that ARM in general right now is a growing market, and servers will be their next push. AMD feels that they need to do something new, and based on ARM growth they think this will be the emerging market that pays off for them.

I'm not saying there are no applications, but those applications are extremely specialized and not representative of the whole market (or even a large section of it). I'm a big proponent of realizing that every tool has its place, but the place of an ARM server, with the current performance disparity and inertia of proprietary code, is a small niche.

If the power/W as performance scales up can remain good, and ARM can outperform a GPU cluster (I'm not convinced, a GPU is massive performance in a small price), I can see a space in HPC, but it is by no means a given.

 

[AtenRa]

The IDC forecast is valid, but then you post AMD marketing spin on it. The message is worthless. AMD is the future because AMD says so? How long have they been saying so? Right now, the question is how long will they survive, not will they become dominant anywhere.

You really believe that because AMD says so that they are well positioned?

You are so blindly anti AMD that you see everything AMD as marketing or propaganda.

First picture shows what the Server forecast is until 2015.

Second picture shows where AMD will provide the server ARM products. It has nothing to do about marketing. Its just shows their server strategy, but you only see marketing.

 

[Idontcare]

I think at this point their end-game no longer involves keeping their company alive. x86 is completely worthless when they fold, so now it's just about generating enough other value (real or imagined) so that the insiders get a nice payday when it's over. Or maybe I'm just too cynical.

I can see this angle too, but I see it as the fall-back angle. If the ARM thing turns into real revenue and profit for AMD, saves their bacon, then they will take that ball and run with it for sure.

But if it doesn't, then at least they will have created some more IP for the holding company to auction off when it comes time to liquidate (or possibly some IP that is enticing enough for a buyer to acquire AMD despite the x86 baggage).

Why is it going to take until 2014 for them to release this?

Today is Oct 31, 2012.

If they taped out an ARMv8 Opteron chip today then the absolute earliest it would be shipping for volumes in production is Q1 2014.

This is an opteron product after all, market is servers, that makes design validation and device verification all the more arduous and necessary.

 

[ShintaiDK]

Posts like that keeps making me thinking of this for some reason:

Looking just at the question of whether a company is acquired and for how much, they find out that golden parachutes work about how you would expect. Companies whose CEOs have golden parachutes are more likely to get acquisition offers and are more likely to be acquired, presumably because their CEOs are les likely to contest takeovers. On the other hand, these companies tend to sell for lower acquisition premiums, again because their CEOs are more likely to be happy to be bought out.

 

[mrmt]

You are so blindly anti AMD that you see everything AMD as marketing or propaganda.

If there is someone blind here it is you, or at least you have some huge memory problems.

According to AMD Bulldozer was the right chip for the cloud, it was THE one. Still can't remember? Here is our friend John Fruehe talking to world + dog about how wonderful Bulldozer is to cloud servers:

http://www.youtube.com/watch?v=rRg2_K5jsC4

What about this white paper here, called "Hybrid Hosting: Evolving the Cloud in 2011. How Superior Price/Performance Can Create a Business Advantage"

http://www.sgi.com/partners/technology/downloads/ADM_Bulldozer_Core_Technology.pdf

So, one year after, Bulldozer is not the right for for the cloud, ARM is the one. Oh, really? So Bulldozer wasn't that marketing, but strategy, or Bulldozer was marketing and with ARM they stopped joking and are now designing credible strategies? But what is AMD doing now? Is strategy to take a growth forecast from another company and put it in your slide as if you have the right product mix to grow with the market? Is strategy to call an event and to announce that in two years you will be shipping generic cores of an unproven technology on the server market?

Who is saying that ARM is the right solution for cloud servers? Who is saying that even if ARM becomes relevant on servers, that AMD servers will be relevant too? That's right, it is AMD executive team and its marketing department.

Btw, you should really take AMD "strategy" slides since 2005 and double check with reality. You'll have quite a few surprises I assure you.

 

[NostaSeronx]

ARM superior Performance/Power and you get more cores...
AMD64 superior Performance/Price and you get less cores...

 

[beginner99]

I'm not an expert but isn't the problem an ARM server is trying to solve already solved with virtualization? (+ the fact that virtualization adds some more benefits).

I don't see the advantages of 8 slow cores over 4 cores with double the speed. Even if you have 8 concurrent requests total time will be the same but with the fast cores average response time is smaller.

What I mean is that the internal applications I'm responsible for while "enterprisey" (=oracle) are all very low volume, eg not much requests but an actual one should have a fast response time (and is rather complex) so a slow ARM thingy would be a no go. On the other hand you don't want your expensive server/cpu idling all the time. So load it full with similar applications -> virtualization.

Servers for webhosting? Well that just screams low margin niche segment.

What the need is an actual APU were the iGPU is more than just a GPU. So HSA thingy makes lot more sense to me just that it will be few years late. Intel was first to fully integrate GPU and it is now still more integrated than AMDs "APUs". However since HSA is software and considering they haven't been able to coem up with anything workign as good as nvidias optimus over 2 years later, my hopes aren't really that high to so so mildly.

 

[AtenRa]

If there is someone blind here it is you, or at least you have some huge memory problems.

According to AMD Bulldozer was the right chip for the cloud, it was THE one. Still can't remember? Here is our friend John Fruehe talking to world + dog about how wonderful Bulldozer is to cloud servers:

http://www.youtube.com/watch?v=rRg2_K5jsC4

What about this white paper here, called "Hybrid Hosting: Evolving the Cloud in 2011. How Superior Price/Performance Can Create a Business Advantage"

http://www.sgi.com/partners/technology/downloads/ADM_Bulldozer_Core_Technology.pdf

So, one year after, Bulldozer is not the right for for the cloud, ARM is the one. Oh, really? So Bulldozer wasn't that marketing, but strategy, or Bulldozer was marketing and with ARM they stopped joking and are now designing credible strategies? But what is AMD doing now? Is strategy to take a growth forecast from another company and put it in your slide as if you have the right product mix to grow with the market? Is strategy to call an event and to announce that in two years you will be shipping generic cores of an unproven technology on the server market?

Who is saying that ARM is the right solution for cloud servers? Who is saying that even if ARM becomes relevant on servers, that AMD servers will be relevant too? That's right, it is AMD executive team and its marketing department.

Btw, you should really take AMD "strategy" slides since 2005 and double check with reality. You'll have quite a few surprises I assure you.

Either you are blind or daft, let me show you the pic again,

Do you NOW see that it says both ARM AND x86 for the Cloud and Mega Data Centers ???
I believe it is now clear to you and everyone else that AMD will not only use ARM CPUs for servers. They will provide ARM CPU Dense Fabric based products for specific workloads for the Cloud and Mega Data Centers.

They will still use Bulldozer based architecture SteamRoller and Excavator in 2013 and 2014 for the server market both in APUs and standalone chips.

ps: lets see some Intel strategies, Itanium, Larrabee, Ultra Thins etc. Just because a strategy doesnt work as they were expecting, it doesn't always making it marketing BS.

 

[MisterMac]

Either you are blind or daft, let me show you the pic again,

Do you NOW see that it says both ARM AND x86 for the Cloud and Mega Data Centers ???
I believe it is now clear to you and everyone else that AMD will not only use ARM CPUs for servers. They will provide ARM CPU Dense Fabric based products for specific workloads for the Cloud and Mega Data Centers.

They will still use Bulldozer based architecture SteamRoller and Excavator in 2013 and 2014 for the server market both in APUs and standalone chips.

ps: lets see some Intel strategies, Itanium, Larrabee, Ultra Thins etc. Just because a strategy doesnt work as they were expecting, it doesn't always making it marketing BS.

Oh wow, now that they offer both dense ARM boxes and their already FAILING SERVER Sku's - everyone is going to buy both?

Really?

Are you that blind?


Server marketshare for Opteron is DECREASING.
Infact without their SuperCompute contracts - try looking at their marketshare.

Just for fun, run the numbers

 

[AtenRa]

AMDs server shipments are fluctuating between 200 to 290 thousand units per quarter for the last 2 years.

They targeting the Cloud and Data centers with the ARM based products, lets see if that strategy will have any valuable effect in there server market share and revenue after 2014.

 

[mrmt]

Do you NOW see that it says both ARM AND x86 for the Cloud and Mega Data Centers ???

I believe it is now clear to you and everyone else that AMD will not only use ARM CPUs for servers. They will provide ARM CPU Dense Fabric based products for specific workloads for the Cloud and Mega Data Centers.

You cannot get a marketing spin even when you are in front of one.

Before Bulldozer launch AMD as touting Bulldozer for everything. Don't forget John Fruehe. After Bulldozer launched, Bulldozer wasn't for everything straight away but could be with a few tweaks. Take for example this title:

Hybrid Hosting: Evolving the Cloud in 2011. How Superior Price/Performance Can Create a Business Advantage

AMD is essentially saying:

"Yes, Bulldozer isn't the best performance or the best performance per watt for the cloud, but once you talk to our sales representative and he gives you a special discount, you will have the best deal of the market".

Now with ARM they are saying:

"Well, we didn't have the right product mix before and you were right, Bulldozer isn't the answer for everything, but in 2014 with ARM you can do plenty of new things with us"

But well, I'm not sure how you are going to spin it, but this is essentially marketing, this is essentially AMD saying to customers how they are going to position themselves.

ps: lets see some Intel strategies, Itanium, Larrabee, Ultra Thins etc. Just because a strategy doesnt work as they were expecting, it doesn't always making it marketing BS.

Why not talk about Core Xeon, a market that is growing each generation that passes, and btw, brings a lot of features developed for Itanium on board?

Why not talk about Knight Corner instead of Larrabee?

Why not mention that the Ultrabook push will at worse enable mainstream tablets with the Core architecture by 2014?

Intel is a big company, they can afford the risk of big projects like that. Some will be a success, like Core or Xeon EX, others won't.

 

[mrmt]

AMDs server shipments are fluctuating between 200 to 290 thousand units per quarter for the last 2 years.

They targeting the Cloud and Data centers with the ARM based products, lets see if that strategy will have any valuable effect in there server market share and revenue after 2014.

In case you don't remember, AMD server sales are losing both in volume and ASP. Situation is not stable, but plunging, as you can read in their Q2 report.

 

[Keysplayr]

Where can I read AMD's Q2 report, mrmt. I would like to see if the numbers jive with what AtenRa just pitched.

 

[AtenRa]

 

[MisterMac]

Take out the Contracts from Cray\Similar - then look at the numbers Q for Q last 1½ years.

 

[AtenRa]

Why take out the Cray ??? It is illegal to sell to Cray ???

 

[Keysplayr]

That, and I expect Q3 and Q4 have seen, and are seeing, dramatic drops.

 

[Keysplayr]

Why take out the Cray ??? It is illegal to sell to Cray ???

Because it's a one off.

 

[AtenRa]

Because it's a one off.

Didnt get that, could you explain ??

 

[Keysplayr]

Didnt get that, could you explain ??

Doesn't really matter or make any difference. What would matter though would be to see Q3 and Q4 of this year and what's going on there. IMHO, if they were selling well, the stock wouldn't be at an all time low right now and falling. It sits at 2.05 right now. Is it under two dollars per share or under 1 dollar per share that analysts usually classify a stock as "junk"?
Not sure what the requirements are for that rating.

 

[mrmt]

Where can I read AMD's Q2 report, mrmt. I would like to see if the numbers jive with what AtenRa just pitched.

You can get data from AMD here:

http://ir.amd.com/phoenix.zhtml?c=74093&p=quarterlyearnings

Here are some excerpts from the transcript:

In Q2:

"Now, turning to our server business, revenue increased from the year-ago period but declined sequentially and was below our expectations. With the introduction of our Bulldozer-based Opteron processors at the end of last year, we saw strong initial adoption in the high-performance compute end market. In the second quarter we experienced a pause in our server business and our focus has to be on building similar acceptance with mainstream IT buyers. Bulldozer is a solid technology, and as we have discussed previously, our growth opportunities are in the portions of the market where our products deliver clear performance advantages. Based on the competitive landscape, we believe Bulldozer can drive modest share growth in the near-term."

In Q3:

"Our server processor revenue declined from the prior quarter, mainly due to lower unit shipments and an ongoing mix change away from higher density servers."

To balance your analysis you can use IDC numbers for 2011, which states AMD share of around 5.5% of the market

http://www.idc.com/getdoc.jsp?containerId=prUS23376112

 

[Cerb]

I'm not saying there are no applications, but those applications are extremely specialized and not representative of the whole market (or even a large section of it). I'm a big proponent of realizing that every tool has its place, but the place of an ARM server, with the current performance disparity and inertia of proprietary code, is a small niche.

If the power/W as performance scales up can remain good, and ARM can outperform a GPU cluster (I'm not convinced, a GPU is massive performance in a small price), I can see a space in HPC, but it is by no means a given.

The same is true for FOSS, as well, but too a lesser degree (it's more a risk issue than one of flat not running). Nothing gets tested as much as default build options on IA32, and lately, x86-64, and both C and C++ easily allow programmers to accidentally shoot themselves in the foot when it comes to an ISA they aren't testing. Given its relative hegemony, you could probably argue that MIPS gets more testing than ARM, too. So even without the code tying one down to an ISA, there is much plenty of inertia, at least for the moment.

ARM superior Performance/Power and you get more cores...
AMD64 superior Performance/Price and you get less cores...

Do we yet have a non-rigged test that shows either one of those to be true? As far as we know it, today, Opterons could very well offer superior performance, price, and power consumption, even for where people may think they want to use these ARM systems (though that's not good for AMD, as it leads into a question of Xeon performance v. Opteron ).

ps: lets see some Intel strategies, Itanium, Larrabee, Ultra Thins etc. Just because a strategy doesnt work as they were expecting, it doesn't always making it marketing BS.

But, Intel can afford to use Benjamins to power the grills at a company cook-out, and still rake in high profits. AMD had a chance, and largely let it pass by targeting high speeds with BD. If they had made the thing to run at 2-3GHz, instead of 4-5GHz, and had actual IPC improvements over Stars, it would have been a great CPU for 'cloud' and virtual hosts, and a fair notebook CPU, to boot. Now, they either have to make it work, push Bobcat-type cores like nobody's business, or we'll find out that the whole BOD is set up with golden parachutes and no liability.

 

[ViRGE]

It looks like JP Morgan is full of downers.:|

http://online.barrons.com/article/SB50001424052748703628504578090810294102342.html?mod=BOL_da_hr

We believe AMD can achieve its embedded segment target of 20% of revenue by fourth-quarter 2013 driven by design wins in the gaming end-market, but we do not expect AMD to gain share in the Server market due to inferior products relative to Intel and a crowded marketplace.

AMD's micro-processing unit (MPU) market share declined roughly 270 basis points over the last four quarters to 16.1% in the third quarter, its lowest level since 2004.

Despite its recently announced restructuring plan and business model, we expect AMD to remain unprofitable in 2013 due to lower revenue and gross margins.

Unprofitable for a whole year is a very bad place to be in for a company as poor as AMD.

 

[ShintaiDK]

Yep, it seems unlike VIA, AMD wants to go out with a bang.